The goal of the proposed research is to explore how infants develop their first lexicon. Previous research on this topic has focused almost exclusively on lexical development after children begin to talk. However, months before this milestone, infants are actively engaged in processing the sounds of their environment to discover the characteristics of words in their language. As early as 8 months of age, infants possess remarkably sophisticated abilities to encode and retrieve words in fluent speech. They also show an astute sensitivity to the properties of speech they hear and the capacity to make predictions about the forms that words in their language are likely to assume. This knowledge is critical to the construction of a lexicon, as learners have to establish which of the many qualities that may characterize any given word in fluent speech should be stored in memory as defining features of the word. This is not a simple challenge for infants who do not know the meanings of most words they hear by 8 months, nor do they know the ways in which their particular language exploits sound to communicate meaning. Therefore, they must rely heavily on cues in the speech they hear to isolate particular details that surface reliably across instances of a word. These details are therefore likely to represent relevant characteristics of a word. Previous research has shown that infants by 8 months do rely heavily on the input for these types of statistical cues to lexical relevance, operating under the principle that details that surface reliably across encounters of words are judged to be invariant by infants and therefore, crucial to lexical identity. Moreover, infants alter their decisions about which cues to preserve in memories for words depending on which cues are statistically favored in the task at hand. The proposed program of research aims to characterize the limits on infants' reliance on statistical cues of the input. Of particular interest are the conditions under which infants selectively make use of statistical information in word discovery and when statistical cues are overridden by other sources of knowledge. Using the Headturn Preference Procedure, word recognition will be measured in infants at different stages of development to characterize the trajectory of early lexical development. By manipulating the types of information (perceptual, statistical and semantic) available to the learner, it is possible to formulate a causal account of how infants successfully construct and refine a lexicon from highly complex and variable input.